Out of memory

[manmanjun]

I use two titanx gpus.Can you tell me the type of GPU you use？

[layumi]

Hi @manmanjun I mainly use the 3x P5000 with 16GB memory. For the stage-II, the input size is small. I use the 3x 2080TI with 11GB.

[manmanjun]

Hi, in the paper, Rectifying Pseudo Label Learning via Uncertainty Estimation for Domain Adaptive Semantic Segmentation, I have a question about the last picture: Prediction Confidence in Fig.5. What is the way to get confidence ?

[layumi]

Hi @manmanjun It is the prediction score after softmax function.

[KaiiZhang]

Hi, @layumi Thanks for your code! I use four 2080Ti gpus, which memory is only 11GB. So in the Stage-I, i may have to set crop-size to 512,256(It is being trained now). I am curious whether high-resolution training images are critical at this stage？

[layumi]

@Zachzhang97 Yes. For the first stage, the resolution of the input image is really important.

[KaiiZhang]

Sounds sad……Do you have any suggestions for reproducing your project on the GPU with smaller memory？

[manmanjun]

Sounds sad……Do you have any suggestions for reproducing your project on the GPU with smaller memory？

Apex is all you need.

[layumi]

@Zachzhang97 Yes. You may try float16, which will affect the performance but is way better than the small input resolution.

[KaiiZhang]

@Zachzhang97 Yes. You may try float16, which will affect the performance but is way better than the small input resolution.

All right, thanks a lot.

[KaiiZhang]

Sounds sad……Do you have any suggestions for reproducing your project on the GPU with smaller memory？

Apex is all you need.

Sorry，i don't understand what you mean.

[KaiiZhang]

Sounds sad……Do you have any suggestions for reproducing your project on the GPU with smaller memory？

Apex is all you need.

Oh, I understand what you mean, thanks !